FLOW CONTROL ALGORITHMS IN NETWORKS WITH VIRTUAL CHANNELS

Abstract

The paper presents a modified approach to flow control in computer networks that support virtual connections by introducing a resource-aware coordination mechanism for bundles of virtual channels. A bundle of virtual channels is defined as a set of co-directed physical links over which multiple software applications establish logical communication paths. Such an abstraction enables concurrent transmission of heterogeneous data streams while sharing a common physical infrastructure. The study analyzes two alternative transmission architectures. In the first scenario, each data stream is mapped to an independent logical TCP connection, which simplifies congestion isolation but increases protocol overhead and resource fragmentation. In the second scenario, a single physical TCP connection encapsulates multiple virtual channels, reducing control overhead and improving link utilization at the cost of more complex internal flow coordination.

The key control variables in a virtual channel bundle are the arrival intensity of data transfer requests and the limited network communication resources, including bandwidth, buffer capacity, and scheduling time. The paper formulates the problem of conflict-free coordination between these parameters as a multi-objective optimization task. A mathematical model of flow control is developed that captures the stochastic nature of request generation, the shared resource constraints, and the presence of competing performance goals—maximization of aggregate throughput and minimization of the average packet delay across all virtual channels. The model represents the network as a distributed system of interacting queues with adaptive service rates and introduces cost functions that reflect the resource expenditure per transmitted data unit.

Based on this model, decentralized algorithms are proposed for request shaping, rate adaptation, and dynamic allocation of communication resources among virtual channels. The algorithms operate using local state information and limited signaling, which ensures scalability and robustness in large-scale distributed environments. A mechanism for balancing the request intensity with the available network capacity is derived, allowing the system to converge to an equilibrium operating point that prevents persistent congestion while maintaining high utilization. Special attention is given to the minimization of transmission cost per virtual channel, achieved through proportional resource distribution and adaptive throttling of aggressive flows.

To evaluate the quality of resource sharing, several fairness criteria are formulated, including max-min fairness, proportional fairness, and Pareto optimality. The Pareto-efficient allocation is shown to provide a balanced trade-off between efficiency and equity, ensuring that no virtual channel can improve its performance without degrading another. The proposed approach enables efficient multiplexing of multiple data streams over shared TCP transport, reduces protocol overhead, and improves delay-throughput characteristics. The results demonstrate that decentralized flow control with virtual channel bundling provides a flexible and scalable solution for modern software-defined and overlay networks where heterogeneous traffic demands must be coordinated over constrained communication resources.